Thermodynamic theory of voting and EU elections

What the math gives to ML

The paper's transferable mechanism is constraint-driven condensation: maximizing entropy for coupled modes while conserving total occupation (probability norm) and mean energy produces a Rayleigh-Jeans distribution, with excess mass accumulating in the lowest-energy mode when excited modes cannot absorb the conserved mass. This can be transferred to mixture-of-experts routing by treating expert loads as occupations and assigning each expert a measurable energy such as latency, loss, or specialization cost. A temperature and chemical-potential controller then creates a tunable transition between diversified routing and deliberate expert condensation. The sharp prediction is that router occupancies should fit the Rayleigh-Jeans law and develop a low-energy condensate when the imposed mass and energy constraints cross a calculable threshold.